The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses strong instructional framing within generated artifacts to ensure future agent sessions strictly adhere to the project's standardized rules and routing logic. For example, the instructions for generating CLAUDE.md include directives such as "CRITICAL — NO EXCEPTIONS" and "No skip. No defer." to enforce the use of AGENTS.md. This behavior is the primary intended function of the harness tool to maintain development consistency.
[EXTERNAL_DOWNLOADS]: The skill references and installs developer tools from trusted and well-known sources. It fetches the skill-creator tool from the official Anthropics skills repository on GitHub and utilizes the @nextstage-brasil/harness CLI for project synchronization and preparation. These downloads are documented parts of the skill's setup process.
[COMMAND_EXECUTION]: The instructions direct the agent to execute shell commands for standard project maintenance tasks. These include git worktree for environment isolation, docker exec for executing backend tests in isolated containers, and npx for running vendor-specific management tools. All commands are scoped to the intended functionality of a project harness.
[INDIRECT_PROMPT_INJECTION]: The skill processes repository source code during its "reverse-engineering" and "mapping" phases, creating an attack surface for indirect prompt injection. The skill mitigates this risk by employing structured templates, reconnaissance checklists, and "anti-leakage" protocols to ensure that external data is processed into sanitized business-level specifications without executing unintended instructions.

ns-harness